Researchers used AI to chain three Chrome flaws into sandbox escape
Security researchers at Xbow used OpenAI’s GPT-5.6 Sol model to help combine three previously unknown Chrome vulnerabilities into an exploit that could run code outside the browser’s sandbox on Windows. Google has fixed the flaws and awarded the team its top Chrome full-chain bounty of $250,000; the case illustrates how AI tools may speed up complex vulnerability research.
The team spent six days on the work and about $40,000 on AI token charges. The exploit chained a bug in Chrome’s Maglev compiler, an animated-WebP decoder flaw and a memory-safety bug in Crashpad to cross both the V8 and operating-system process sandboxes. Xbow estimated around one billion users could be affected; Chrome users should update to version 152.0.7977.82 or later and relaunch the browser for the fix to take effect.
- Xbow chained three Chrome zero-days with AI assistance.
- Google paid the team a $250,000 bounty.
- Update Chrome to version 152.0.7977.82 or later.
New here? Start with this
Chrome is Google’s web browser, and a security flaw is a mistake in its code that could let an attacker do something the software was meant to prevent. Browsers use sandboxes to limit what a webpage or browser process can access, helping contain the damage if a flaw is exploited.
Xbow is a security research team that investigates software weaknesses. Researchers reported combining three flaws in different parts of Chrome so that, together, they could get past the browser’s safeguards and run code outside its sandbox on Windows.
The case matters because it shows how several separate weaknesses can be linked to overcome layers of protection, and how AI tools may assist that research. Google’s fixes address the reported flaws; Chrome users need an updated browser for those fixes to take effect.
AI Research Science Technology
Read the full article at the source →
Originally published by The Register as “Security researchers spent 6 days doing their Chromework, built 1B user exploit”.